When psychometrics meets epidemiology, and the studies which result! Tom Booth [email protected] Lecturer in Quantitative Research Methods, Department of Psychology, University of Edinburgh Who am I? • MSc and PhD in Organisational Psychology – ESRC AQM Scholarship • Manchester Business School, University of Manchester • Topic: Personality psychometrics • Post-doctoral Researcher • Centre for Cognitive Ageing and Cognitive Epidemiology, Department of Psychology, University of Edinburgh. • Primary Research Topic: Cognitive ageing, brain imaging. • Lecturer Quantitative Research Methods • Department of Psychology, University of Edinburgh • Primary Research Topics: Individual differences and health; cognitive ability and brain imaging; Psychometric methods and assessment. Journey of a talk…. • Psychometrics: • Performance of likert-type response scales for personality data. • Murray, Booth & Molenaar (2015) • Epidemiology: • Allostatic load • Measurement: Booth, Starr & Deary (2013); (Unpublished) • Applications: Early life adversity (Unpublished) • Further applications Journey of a talk…. • Methodological interlude: • The issue of optimal time scales. • Individual differences and health: • Personality and Physical Health (review: Murray & Booth, 2015) • Personality, health behaviours and brain integrity (Booth, Mottus et al., 2014) • Looking forward Psychometrics My spiritual home… Middle response options Strongly Agree Agree Neither Agree nor Disagree Strong Disagree Disagree 1 2 3 4 5 Strongly Agree Agree Unsure Disagree Strong Disagree 1 2 3 4 5 Yes ? No 1 2 3 Middle response options • What we assume… • That it represents the mid-point of the latent trait standing. • How they are used… • Highly variable • Could represent the mid-point, uncertainly, would prefer not to say, don’t understand the question… • A number of studies have looked at the issue, but none have done so: • In a large representative sample • Using an entire omnibus measure of personality • With optimal methods • And considered the impact of practical scoring algorithms What we did… • Data: • US standardization sample of the 16 Personality Factor Questionnaire, version 5 (16PF5; n=10,261) • UK standardization sample of the 16PF5 (n=1,212) • Analysis: • Approximate uni-dimensionality based on single factor CFA’s • Compared fit of Nominal Response Model (NRM; Bock, 1972) to the Graded Partial Credit Model (GPCM; Muraki, 1992) • Investigated category and threshold ordering based on NRM Ideal item response curves Category disordering Threshold disordering US Results UK Results Example Items Item 1 of Sensitivity (I) in US sample Item 1 Openness to Change (Q1), US sample Practical Impact Scores standardized for comparison: NRM score <-> Sum; r = .83 NRM score <-> Missing data; r = .77 NRM score <-> Collapse with top; r = .62 NRM score <-> Collapse with bottom; r = .95 Practical Impact Epidemiology (or something quite similar!) Introducing the Lothian Birth Cohort 1936 • LBC1936 is a longitudinal study of ageing. • Surviving members of Scottish Mental Survey 1947. • Wave 1 (n=1091), mean age 69.5 (SD 0.8) • Medical history, psychometric measures, cognitive testing, physical exam, blood samples. • Wave 2 (n=866), mean age 72.5 (SD 0.7) • Repeat testing from wave 1. Additionally underwent brain scans. • For full sample details see (Deary et al. 2007; Deary et al. 2011; Wardlaw et al. 2011). Allostatic load and stress response • When completing my post-doc in CCACE published on AL. • Allostasis: Maintaining stability through change. • Homeostasis mentioned by Ram et al. • Allostatic Load: Physiological cost of exposure to a stressor. (Ganzel et al. 2010; McEwen & Wingfield, 2003). AL in the literature • Del Giudice et al. (2012) – “Adaptive Calibration Model” • Chen & Miller (2012) – “Shift-and-Persist” • Common features: • Key developmental periods in defining biological response to stress. • Early life stress (low SES, Adversity etc.) may be a major source of allostatic load. • Key roles of the sympathetic, parasympathetic and HPA-axis systems. • Individual differences may act as a buffering mechanism. • Effects of allostatic overload seen in numerous biomarkers including brain tissue integrity. AL in the literature • SES/Environments -> Health & Biomarkers of Allostatic Load • E.g. Dowd et al. (2009); Szanton et al. (2005); Seeman et al. (2010) • Biomarkers of AL -> Health: • E.g. Carlson & Chamberlain (2005); Juster et al. (2010) • Importance of key developmental periods: • E.g. Wilkinson & Goodyear (2011); Howell & Sanchez (2011) • Role of personality and individual differences: • E.g. Chapman et al. (2009); Nabi et al. (2008) Measuring AL • Across studies (e.g. Szanton et al. 2005; Chen et al. 2012) suggest 8 areas and markers: 1. Sympathetic Nervous System: Nor/Epinephrine 2. Parasympathetic Nervous System: High/low frequency heart rate 3. HPA-Axis: Cortisol, DHEA. 4. Inflammation: C-reactive Protein, Iterleukin-6, fibrinogen. 5. Cardiovascular: Systolic and diastolic BP, heart rate. 6. Glucose Metabolism: Glucose, Insulin resistance. 7. Lipid Metabolism: High/low-density lipoprotein, BMI, waist-hip ratio, triglycerides. 8. Organ function: Creatinine, Respiratory peak flow. • Seeman et al. (2010) produced a second order CFA model using 5 areas as factors and 17 biomarkers. Replicating the measurement model • Based on wave 2 data from the LBC1936 • Allostatic load measures: • Metabolic Markers: Triglycerides (Trig), High density Lipoprotein (HDL), low density lipoprotein (LDP), HBa1c & Body-Mass Index (BMI). • Inflammation Markers: Fibrinogen, C-reactive Protein (CRP) & Interlukin-6 (IL6) • Blood Pressure: Mean sitting systolic and diastolic blood pressure (SBP & DBP), & ortostatic hypotension (orto). Replicating the measurement model Fibrinogen 0.59 CRP 0.83 Inflam. IL6 0.69 Trig 0.39 0.45 -0.37 HDL -0.40 -0.02 LDL -0.20 Metab. HBa1c 0.42 BMI 0.60 -0.07 Diastolic BP 0.57 BP Systolic BP 0.95 Booth, Starr & Deary (2013). Modeling multisystem biological risk in later life: Allostatic load in the Lothian birth cohort study 1936. American Journal of Human Biology, 25, 538-543 Replicating the measurement model Fib. 0.55 CRP 0.82 Inflam. IL6 0.74 0.43 Trig 0.35 -0.39 HDL -0.39 LDL -0.24 Metab. 0.87 AL HBa1c 0.43 0.71 BMI 0.61 Hyp. Booth, Starr & Deary (2013). Modeling multisystem biological risk in later life: Allostatic load in the Lothian birth cohort study 1936. American Journal of Human Biology, 25, 538-543 Replicating the measurement model 0.13 Fibrinogen 0.62 0.35 CRP 0.74 Inflam. 0.40 IL6 0.58 0.33 Trig 0.57 AL -0.37 HDL -0.62 -0.31 LDL 0.14 Metab. 0.43 HBa1c 0.04 0.59 BMI 0.09 0.50 Hyp. Booth, Starr & Deary (2013). Modeling multisystem biological risk in later life: Allostatic load in the Lothian birth cohort study 1936. American Journal of Human Biology, 25, 538-543 Measuring AL: Factor or composite? Common methods of assessment • Sample specific cut-points (deciles, quartiles) • Clinical cut-points • Composite scores (summed z-scores, PCA) • Latent models (higher-order, bi-factor) Measurement of AL AL-bifactor AL-PCA AL-z AL-dec AL-quart AL-clin AL-bifactor 1.00 AL-PCA 0.58 1.00 AL-z 0.47 0.72 1.00 AL-dec 0.39 0.65 0.59 1.00 AL-quart 0.45 0.69 0.67 0.68 1.00 AL-clin 0.42 0.65 0.80 0.60 0.78 1.00 Booth (Unpublished). Supplementary allostatic load analyses in the LBC1936. AL and early life adversity AL-bifactor AL-PCA AL-z AL-dec AL-quart AL-clin Adversity 0.09 0.09 0.10 0.12 0.11 0.04 0.11 0.08 0.09 0.10 0.11 0.02 • Childhood Adversity (Johnson et al. 2010) scores based on living environment at age 11 years: • People/rooms in house ratio. • People per toilet. • Indoor toilet (yes/no). • Fathers occupational class. Booth (Unpublished). Supplementary allostatic load analyses in the LBC1936. Other AL Studies • Medication of AL-> cognitive ability association by brain volumes • AL white matter tract integrity associations using TBSS: assessing localization and lateralization of effects. • Moderated mediation of depression and AL on cognitive decline by APOE status • AL and methylation age • Optimal measurement of AL by Booth et al. (2015). Association of allostatic load with brain structure and cognitive criterion association ability in later life. Neurobiology of Ageing, 36, 1390-1399 A methodological interlude The importance of time when integrating social and biological Multiple timeframe studies of biosocial processes • The broad argument is seated in what might now be considered classical distinctions for longitudinal studies (Nesselroade, 1991): • Change: Longer term processes usually studied within traditional panel or cohort studies. • Variability: Shorter term processes or fluctuations, now becoming studied with experience sampling methodologies. • But what happens when we extend to multiple timeframes, and pose hypotheses about the relations between effects at these different time frames? AL and multiple time frames Process Time scale Allostatic load (accumulation) Life long Environmental influence Life long but periodic – we move, situations change Traits (personality, IQ, anxiety) Argued long term stability, but evidence of high within person variability Biomarkers Momentary response cycles (cortisol) Optimal time lags "With impoverished theory about issues such as when events occur, when they change, or how quickly they change, the empirical researcher is in a quandary. Decisions about when to measure and how frequently to measure critical variables are left to intuition, chance, convenience, or tradition. None of these are particularly reliable